contours(return_type="geopandas") raises ValueError on empty result with a CRS

[brendancol]

Describe the bug

contours(agg, return_type="geopandas") raises a ValueError when the input raster has a crs attribute but the requested levels produce no contour lines. This happens for a flat raster, for levels outside the data range, or any case where the result set is empty.

The error comes from _to_geopandas() in xrspatial/contour.py:

gdf = gpd.GeoDataFrame(records, crs=crs)

When records is empty and crs is not None, geopandas raises:

ValueError: Assigning CRS to a GeoDataFrame without a geometry column is not supported

A related problem: the all-NaN early-return path calls _to_geopandas([], None), which avoids the crash but drops the input CRS. So an empty result either crashes (when a CRS is present) or silently loses the CRS (on the all-NaN path).

Reproduce

import numpy as np, xarray as xr
from xrspatial import contours

flat = xr.DataArray(
    np.ones((4, 4)), dims=["y", "x"],
    coords={"y": np.linspace(2, 0, 4), "x": np.linspace(0, 2, 4)},
    attrs={"crs": "EPSG:5070"},
)
contours(flat, levels=[5.0], return_type="geopandas")  # raises ValueError

Expected behavior

An empty result returns an empty GeoDataFrame that still carries the input CRS and has a geometry column, with no exception.

Additional context

Found by the deep-sweep metadata propagation sweep on the contour module. The existing geopandas test uses a raster with no crs attribute and a non-empty result, so this path is uncovered.